JP2024512037A - How to determine the current glucose level in the transport fluid - Google Patents

Abstract

The present invention is a method for continuously determining current glucose values in a transport fluid, particularly blood, of an organism, comprising: a) temporally spaced tissue glucose values in tissues surrounding said transport fluid; b) determining a measurement noise value from the series of measurements of said sensor based on a measurement model in the form of a linear or non-linear function; c) determining the tissue glucose value taking into account process noise values; and c) a state transition model used to assign the current glucose value in the transport fluid to the determined tissue glucose value taking into account process noise values. d) using a Kalman filter in the case of a measurement model in the form of a linear function or using an extended Kalman filter in the case of a measurement model in the form of a non-linear function; estimating the current glucose value in the transport fluid based on an approximation of the determined tissue glucose value.

Description

The present invention relates to a method for continuously determining the current glucose value in a biological transport fluid, in particular blood.

Furthermore, the invention relates to a device for continuously determining the current glucose value in a biological transport fluid, in particular blood.

The invention also relates to an evaluation device for continuously determining the current glucose value in a biological transport fluid, in particular blood.

Furthermore, the present invention relates to a non-volatile computer readable medium for storing instructions that, when executed on a computer, cause the method to continuously determine the current glucose level in a transport fluid of a living organism, in particular blood. .

Although the invention is applicable to any method for determining current glucose values in a transport fluid, the invention will be described in relation to blood glucose concentrations in living organisms.

Systems for continuous glucose monitoring, also known as CGM--continuous glucose monitoring, have become known for determining the blood glucose concentration BG in living organisms, especially humans.
In a CGM system, interstitial tissue glucose concentration IG is automatically measured, for example, every 1 to 5 minutes.
In particular, diabetic patients can take measurements significantly more frequently compared to self-monitoring procedures, also known as SMBG, where patients manually determine their blood glucose levels 4 to 10 times a day. , benefit from the CGM system.
This allows automatic assessment and warning signals to be sent to the patient while he or she is sleeping, helping to avoid critical health conditions for the patient.

On the one hand, known CGM systems are based on electrochemical processes.
Such a CGM system is described, for example, in WO 2006/017358.
Furthermore, optical CGM systems have become known, for example, from German Patent Application No. 10 2015 101 847, in which glucose value-dependent fluorescence is used, which is incorporated herein by reference.
Both types of CGM systems measure interstitial tissue glucose concentrations.

It is also known that tissue glucose concentration or interstitial glucose concentration IG is different from blood glucose concentration (hereinafter abbreviated as BG).
Non-patent literature “Time lag of glucose from intravascular to interstitial compartments in humans” by Basu, Ananda et al. (Diabetes (2013): DB-131132) as described, e.g. through the intake of food or nutrients, or , upon administration of insulin, there is a large deviation after a strong effect on blood glucose levels.
This shift is caused by diffusion processes in the tissues surrounding the blood, so that the IG values follow the BG values with a time delay and are attenuated, as described, for example, in Rebrin, Kerstin et al. “Subcutaneous glucose predicts plasma glucose independent of insulin: implications for continuous monitoring” (American Journal) of Physiology-Endocrinology and Metabolism 277.3 (1999): E561-E571).

Due to the decay and time delay mentioned between the two glucose concentrations, on the one hand in the blood BG and on the other hand in the surrounding tissue IG, for example, the blood glucose concentration due to a drop of blood extracted from a finger and a drop Calibration of CGM systems using manual determination of glucose concentrations in the blood, determined using external measurement devices, results in significant inaccuracies.

However, in order to achieve accurate calibration of the CGM system, the aforementioned differences between tissue and blood glucose concentrations must be taken into account or estimated.
Various methods have become known for this purpose. Non-Patent Literature Keenan, D. Barry et al., “Delays in minimally invasive continuous glucose monitoring devices: a review of current technology.” (2009): 1207-1214) using a time-delayed glucose signal for calibration. It is known.
Furthermore, non-patent literature Knobbe, Edward J. and “The extended Kalman filter for continuous glucose monitoring.” by Bruce Buckingham et al. 5): From 15-27), the decay and time of the diffusion process of glucose between blood and tissue by Kalman filter It is known to compensate for delays.

However, the problem here is that mobile devices for continuous determination of current glucose values have limited computing and energy resources.
To determine the current glucose value, only a relatively small amount of computing power is available, coupled with limited energy, and as a result, previously known methods cannot be implemented on mobile devices; Or it can only be performed for a short time, severely limiting its usefulness.

International Publication No. 2006/017358 pamphlet German Patent Application No. 10 2015 101 847

“Time lag of glucose from intravascular to interstitial compartments in humans” by Basu, Ananda et al. (Diabetes (2013): DB-131132) Rebrin, Kerstin et al., “Subcutaneous glucose predicts plasma glucose independent of insulin: implications for continuous monitoring” (American Journal of Physiology-Endocrinology and Metabolism 277.3 (1999): E561-E571) Keenan, D. Barry et al., “Delays in minimally invasive continuous glucose monitoring devices: a review of current technology.” (Journal of Diab etes Science and Technology 3.5 (2009): 1207-1214) Knobbe, Edward J. and Bruce Buckingham et al., “The extended Kalman filter for continuous glucose monitoring.” (Diabetes technology & therapy 7.1 (2005) ):15-27)

The aim of the invention is therefore to identify a method, a device and an evaluation device that allow a more accurate determination of glucose values in the blood with fewer resources and simpler implementation.

A further object of the invention is to identify alternative methods, alternative devices and alternative evaluation devices.

A further object of the invention is to provide a device and an evaluation device with improved determination of blood glucose concentration in an organism based on measurements of interstitial tissue glucose concentration.

The present invention solves the above-mentioned object by a method for continuously determining the current glucose value in a transport fluid of an organism, in particular blood, which method comprises:
a) using a sensor to measure a series of at least two measurements of temporally spaced tissue glucose values in tissue surrounding the transport fluid;
b) determining the tissue glucose value from the series of measurements, taking into account measurement noise values, based on a measurement model in the form of a linear or non-linear function;
c) providing a state transition model used to assign current glucose values in the transport fluid to the determined tissue glucose values taking into account process noise values;
d) based on the approximation of said state transition model and said tissue glucose value using a Kalman filter in the case of a measurement model in the form of a linear function or using an extended Kalman filter in the case of a measurement model in the form of a non-linear function; estimating the current glucose value in the transport fluid.

Furthermore, the invention provides a method for continuously estimating the current glucose value in a transport fluid of an organism, in particular blood, suitable for carrying out the method according to any one of claims 1 to 15. achieve the above objectives by means of a device for
a sensor for measuring fluorescence in the tissue surrounding the transport fluid by means of a probe, in particular a polymeric fiber optic probe, designed to measure a series of measurements;
at least two temporally spaced measurements of tissue glucose values in tissue surrounding the transport fluid;
A providing apparatus designed to provide a state transition model for assigning a glucose value in a transport fluid to a determined tissue glucose value taking into account a process noise value; a providing device used to provide a measurement model in the form of a linear or non-linear function, wherein the measurement model providing the measurement from the tissue glucose value of the sensor is assigned taking into account a measurement noise value;
Based on the measurement model, the determined set of measurements is used to determine the tissue glucose value, and also using a Kalman filter in the case of a measurement model in the form of a linear function, or using a Kalman filter in the case of a measurement model in the form of a linear function; Designed to estimate the current glucose value in the transport fluid based on the provided state transition model and an approximation of the determined tissue glucose value, using an extended Kalman filter in the case of the morphology measurement model. and an evaluation device.

Furthermore, the present invention achieves the above-mentioned object with an evaluation device for continuous estimation of the current glucose value in a transport fluid of a living organism, in particular blood;
an interface for connecting a sensor for providing a series of measurements, the interface comprising at least two measurements of tissue glucose values in tissue surrounding the transport fluid;
It is used to assign the tissue glucose value determined by the state transition model to the glucose value in the transport fluid, taking into account the process noise value, and to store the measurement model in the form of a linear or nonlinear function, the measurement model a memory used to assign sensor measurements to tissue glucose values, taking into account measurement noise values;
Based on the measurement model, a series of measurements is used to determine the tissue glucose value, and also using a Kalman filter in the case of a measurement model in the form of a linear function, or a measurement in the form of a non-linear function. and a computing device designed to estimate a current glucose value in the transport fluid based on the state transition model and the approximation of the tissue glucose value using an extended Kalman filter in the case of the model.

Furthermore, the present invention provides a method for carrying out the present invention in biological transport fluids, in particular blood, which, when executed on a computer, is suitable for carrying out the method according to any one of claims 1 to 15. Achieving the above objects by a computer readable medium for storing instructions for carrying out a method for continuous estimation of glucose values;
a) using a sensor to measure a series of at least two measurements of temporally spaced tissue glucose values in tissue surrounding the transport fluid;
b) determining the tissue glucose value from the sensor series of measurements based on a measurement model in the form of a linear or non-linear function, taking into account measurement noise values;
c) providing a state transition model used to assign current glucose values in the transport fluid to the determined tissue glucose values taking into account process noise values;
d) based on a state transition model and an approximation of tissue glucose values using a Kalman filter in the case of a measurement model in the form of a linear function or using an extended Kalman filter in the case of a measurement model in the form of a non-linear function; estimating a current glucose value in the transport fluid.

That is, a method for estimating blood glucose concentration in living organisms has been proposed.
This has the following procedural steps.

In a first method step, a series of measurements is recorded by one or more sensors with at least two temporally spaced sensor measurements of interstitial tissue glucose values in a tissue of the organism.

In a further step, a measurement or sensor model of the relationship between the sensor measurements and the tissue glucose value is provided, and one or more states comprising a model of the relationship between the tissue glucose value and the blood glucose value. A transition model is provided.

In a further method step, the blood glucose value of the organism is quantified by an estimate based on a state transition model and an approximation of the tissue glucose value, the estimate using a Kalman filter in the case of a measurement model in the form of a linear function. Alternatively, in the case of measurement models in the form of nonlinear functions, it is essential to perform using an extended Kalman filter.

The Kalman filter is an unbiased and consistent estimator with minimal variance.
Because of these estimated properties, the Kalman filter is an optimal linear filter.
Similarly, in contrast to other (recursive) linear estimators that minimize least squares, Kalman filters allow processing of problems involving correlated noise components.

The extended Kalman filter is a nonlinear extension of the Kalman filter described above.
The extended Kalman filter analytically approximates a nonlinear problem with a linear problem based on a nonlinear function.

In this case, the evaluation device can be, for example, a computer, an integrated circuit, etc. designed for optimized calculations such as matrix tracing.
The device and/or the evaluation device can be designed as a portable device with an independent energy source that maintains efficient operation, e.g. a battery, a rechargeable battery, thus allowing battery operation for as long as possible. In order to achieve this, the energy consumption for performing the method according to an embodiment of the invention is as low as possible and the user experience is improved.
In particular, power saving processors, circuits, switching circuits, interfaces, in particular wireless interfaces, etc. can be used for this purpose.
Embodiments of the method can be adapted with respect to its parameters, e.g. with respect to the underlying device or evaluation device, e.g. with respect to the evaluation horizon and/or noise horizon, the range of random samples, linear or non-linear functions, etc. This will be explained below in order to achieve sufficient accuracy on the one hand and long execution times on the other hand.

One of the possible advantages that can be achieved with the embodiments is that the current glucose value in the transport fluid, in particular the blood, can be estimated in an efficient manner in terms of time and computer resources.
Another advantage is that there is no restriction on specific sensor models and/or state transition models, which greatly increases flexibility compared to known methods.
A further advantage is that not only the accuracy of the current glucose value is increased, but also the past glucose values are improved at the same time.

Other features, advantages, and other embodiments of the invention are described or will become apparent from the following description.

According to the present invention, a plurality of state transition models are provided that change depending on the course of the estimated current glucose value over time, particularly its rate of change over time.
This allows efficient and at the same time accurate determination of current glucose values.

Furthermore, according to the invention, at least two state transition models are provided, one based on a constant glucose concentration, one based on a constant change in glucose concentration, and/or one based on a constant change in glucose concentration. Based on a weighted sum of previous glucose concentrations.
This allows the state transition model to be adapted based on the dynamics of glucose values, thereby allowing current glucose values to be determined in a resource efficient manner.
This avoids overestimating or underestimating the glucose value in the blood when it rises or falls.

Furthermore, according to the invention, the measured values are filtered by at least one filter function, and the errors of the sensor, in particular the measurement errors, are suppressed by the at least one filter function.
Using the filter function, defective measurements, e.g. sensor errors or outliers in the measurements, can be screened out in a simple way, i.e. they are not taken into account in the further calculation of the current glucose value. .

Furthermore, according to the invention, at least one measurement noise value is adjusted periodically.
This is necessary, on the one hand, to ensure that the current glucose value is sufficiently accurate, and on the other hand, to avoid unnecessary adjustments that do not reflect, or only slightly reflect, an increase in the accuracy of the current glucose value. or ensuring that the respective measurement noise values are adjusted efficiently, in particular at regular intervals, in order to avoid updates.

Furthermore, according to the invention, in order to adapt the at least one measurement noise value, the variance of the measurement noise value is determined using a random sample of the measurement values, and in particular the variance is estimated.
A possible advantage in this regard is that the measurement noise value can be adjusted efficiently.

Furthermore, according to the invention, statistical tests, in particular the Kolmogorov-Smirnov test, are used to determine whether the null hypothesis (that the random sample follows a mean free Gaussian distribution with a determined variance of the measured noise values) is not rejected. It will be checked whether
One possible advantage is that it makes it possible to adapt the measurement noise value in a simple way.

Furthermore, according to the invention, the variance of the measurement noise values is determined for at least one further sample of the measurement values, as long as the null hypothesis is rejected.
A possible advantage is that it is an efficient way to adjust the measurement noise value.

Furthermore, according to the invention, the measured values are checked for outliers using at least one filter function, in particular using the NIS test, and measured values determined as outliers are discarded.
A possible advantage is that the accuracy of the determination of the current glucose value is further improved.

Furthermore, according to the invention, at least one filter function is used to check for measurements exceeding and/or below specified limit values; Value is discarded.
This represents a particularly simple way of checking measurements for outliers.

Furthermore, according to the invention, if at least a predetermined number, in particular two temporally consecutive previous measurement values, have been rejected as measurement errors, the current measurement value, which was not determined as an outlier, nevertheless Therefore, it is rejected as a measurement error.
This means that failures of the measurement system are assumed to be considered terminated only when all conditions or limit values are met.
A possible benefit of this is to further improve the accuracy of determining the current glucose value.

Furthermore, according to the invention, the state transition model includes a diffusion model for time-dependent modeling of the diffusion process of glucose from the transport fluid into the surrounding tissue.
In particular, using diffusion models based on diffusion constants, it is possible to model transport fluids, in particular the decay and time delay between glucose values in blood and tissue glucose values, in a simple and less computationally intensive way. be.

Furthermore, according to the invention, several previous measurements are filtered by a "Kalman fixed interval smoother".
A possible benefit of this is that past readings are smoothed, the rate of change of blood glucose values is improved, and current glucose readings are improved.

Furthermore, according to the invention, if a "Kalman fixed interval smoother" is used, it is performed in forward and reverse directions, with Kalman filtering in the forward pass and RTS filter and/or in the reverse pass. Or an MBF filter is used.
This allows efficient application of the "Kalman fixed interval smoother".
For nonlinear measurement models, RTS filters or "enhanced RTS filters" reduce computational effort when using steady-state transition models.
The MBF filter reduces the computational effort of the unsteady state transition model.

Furthermore, according to the invention, the trends in blood glucose concentration are classified using several categories, in particular using at least seven categories.
Therefore, the trend or future course of blood sugar can be displayed to the user in a simple and efficient manner.

Further important features and advantages of the invention result from the dependent claims, the drawing and the description with reference to the drawings.

It goes without saying that the features mentioned above and those further explained below can be used not only in the combinations specified in each case, but also in other combinations or alone, without departing from the scope of the invention.

Embodiments of the invention are shown in the drawings and are explained in more detail in the description below.
All transformation steps such as equations, assumptions, solution methods, etc. can be used separately without departing from the scope of the invention.

1 schematically depicts the steps of a method according to an embodiment of the invention; FIG. FIG. 3 is a diagram showing the course of blood sugar over time when using a Kalman filter and a Kalman smoother according to an embodiment of the present invention. FIG. 3 is a diagram illustrating the course of blood glucose over time using a Kalman filter and Kalman smoother according to an embodiment of the invention with trend estimation;

FIG. 1 shows that the current glucose value in the transport fluid of the organism, in particular the blood, is determined in particular continuously and, in the case of a measurement model in the form of a linear function, based on the use of at least one Kalman filter, or In the case of a measurement model in the form of a non-linear function, the steps of a method for determining the glucose concentration in blood using at least one extended Kalman filter are detailed.

This procedure includes the following steps.

In a first step S1, a sensor is used to determine a series of measurements comprising at least two temporally spaced measurements of tissue glucose values in tissue surrounding the transport fluid.

In a further step S2, a tissue glucose value is determined using the determined set of measurements based on a measurement model in the form of a linear or non-linear function, the measurement model taking into account at least one measurement noise value. and is used to assign sensor readings to tissue glucose values.

In a further step S3, at least one state transition model is provided, the at least one state transition model adjusting the determined tissue glucose value to the at least one glucose value in the transport fluid, taking into account at least one process noise value. used to allocate.

In a further step S4, the current glucose value is determined using at least one Kalman filter in the case of a measurement model in the form of a linear function or at least one extended Kalman filter in the case of a measurement model in the form of a non-linear function. are estimated based on at least one provided state transition model and an approximation of the determined tissue glucose value.

Further embodiments of the invention will be described in detail with reference to FIGS. 2 and 3, also with reference to the use of different filters, identification of outliers, etc.
The respective features can be combined with each other in whole or in part.

Application of the Kalman filter will be explained below.

Kalman filter is an estimator of dynamic variables with Gaussian distribution measurements and process noise.
Furthermore, a Markov property is required in which each state depends only on its previous state.

は、システム行列Ｆ_ｋ、前の状態ベクトルｘ_ｋおよび定義されたプロセスノイズｗ_ｋによるものである。 arbitrary state vector

is due to the system matrix F _k , the previous state vector x _k and the defined process noise w _k .

測定ノイズの共分散には、以下が適用される。

フィルタは、第１の予測ステップおよび第２のイノベーションステップの２つのステップからなる。 Furthermore, this is the measurement vector

The following applies to the measurement noise covariance:

The filter consists of two steps: a first prediction step and a second innovation step.

２）イノベーションステップ：

1) Prediction step:

2) Innovation step:

Tissue glucose dynamics, ie, changes in sugar concentration in tissues surrounding blood vessels, are influenced by the diffusion of glucose from the blood.
Modeling of blood glucose dynamics is performed by modeling with uncertainty, as control variables such as food intake or insulin concentration are not known.

A simple possibility is to model blood glucose assuming a constant blood glucose concentration.

および

The change in blood glucose concentration is then modeled using the process uncertainty w.
Assuming that disturbances such as increases in blood glucose concentration due to feeding or decreases due to administration/release of insulin affect blood glucose changes, one embodiment of the present invention uses a model based on constant blood glucose changes to calculate changes in blood glucose concentrations. I do.

and

The disturbance is modeled by the process error w of blood sugar change.

をもたらす。

bring about.

If the assumption is that process noise is uncorrelated and white noise is violated, the model can be extended to meet the requirements again.
If the current process noise value correlates with past process noise values, the Markov property is violated.

でもたらされる。 According to the explanation of blood sugar dynamics, changes in blood sugar do not occur suddenly, but this is because the drop or rise in blood sugar becomes noticeable over a certain period of time.

brought about by

が行われる。
現在値は、以前の値の加重和によってモデル化される。
ここで、このモデリングは、各状態がその前の状態にのみ依存する状態のマルコフ特性によるカルマンフィルタの要件と矛盾することに留意されたい。 In a further embodiment of the invention, blood glucose dynamics may be calculated using an autoregressive model (AR model) of order p.

will be held.
The current value is modeled by a weighted sum of previous values.
Note here that this modeling contradicts the requirement of a Kalman filter with Markov properties of states where each state depends only on its previous state.

定血中グルコース変化モデルに対応する。

Corresponds to the constant blood glucose change model.

To model the diffusion process of blood glucose in the blood to the surrounding tissue and then from the tissue to the sensor, in one embodiment of the invention, the diffusion process of glucose from the blood to the tissue and then to the sensor is , is summarized by considering the process as a series circuit.

Thereby, the concentration of glucose in the sensor g ^s and the time constant τ consist of the sum of the time constants of the diffusion of glucose from the blood to the tissue fluid and from the tissue fluid to the sensor.

The time-discrete formula with sampling Δt is obtained by applying the matrix exponential function as follows.

overall state vector

To determine the tissue glucose value, the sensor measurements use a series of measurements based on a measurement model in the form of a linear or non-linear function using the measurement model, taking into account at least one measurement noise value. and assigned to tissue glucose values.
Various measurement models based on linear or non-linear functions are available for this purpose, according to embodiments of the invention.

に基づく非線形測定モデルを使用する場合、拡張カルマンフィルタが使用される。
この目的のために、測定行列Ｈ_ｋは、一次テイラー多項式

によって近似することができ、式中、

および

である。 For a linear measurement model, it involves a sensitivity e and an offset o.

When using a nonlinear measurement model based on , an extended Kalman filter is used.
For this purpose, the measurement matrix H _k is a first-order Taylor polynomial

It can be approximated by, where,

and

It is.

One problem in estimating blood sugar from continuous measurement data, so-called continuous glucose monitoring data (abbreviated CGM data), is that changes in blood sugar due to food intake or the effects of insulin, for example, may occur over a period of 10 minutes. It is only noticeable in interstitial fluids after a time delay of more than 30 minutes.
On the one hand, this is due to physiological reasons.
On the other hand, additional time is required for the interstitial fluid to diffuse into the sensor and be measured.
As a result, when blood sugar increases, the estimate first lags and then rises very rapidly if there is also a change in tissue sugar, which represents a non-physiological behavior.

Over a period of time, depending on the food eaten and the administration of insulin, blood sugar rises almost constantly until insulin takes effect.
This change decreases from there until blood sugar decreases at a nearly constant rate.
In healthy individuals, this process is easily predictable due to the body's control loops.
In the case of diabetics, this cannot be determined without knowing the administration of insulin, i.e., insulin-specific parameters such as time point and effective time, among others.

Another change in blood glucose levels occurs when insulin weakens.
Also, this point can only be determined using additional knowledge regarding the amount of insulin, duration of action, etc.
This results in a dynamic model with constant blood glucose changes, where increasing blood glucose results in a significant overestimation of blood glucose and decreasing blood glucose results in a corresponding underestimation.

The effect can be reduced by changing between dynamic models in a controlled manner.

For example, in one embodiment of the present invention, depending on certain parameters, between a constant rate of change (cROC) model with modeling of uncertainty due to process noise and a constant blood glucose model (cBG). It is possible to switch back and forth.

The cBZ dynamic model is selected if a predefinable hypoglycemic value BZ _u is not reached or if the blood glucose concentration increases (ROC>0) and exceeds the upper blood glucose limit BZ _o .

In one embodiment of the invention, continuous adjustment of the measurement noise, which is assumed to be free of mean values and Gaussian distributions, is performed to further improve the estimation of blood glucose content.
This makes it possible to take into account the dispersion between different sensors and the aging of the sensors.
Adjustment or updating of the variance directly leads to a change, in particular an improvement, in the quality of the estimate regarding the current glucose value.
However, if the measurement noise value is underestimated, this results in a very noisy measurement signal, leading to erroneous measurements.
On the other hand, if the estimated value of the measurement noise value is too high or the estimated value of the process noise is too low, a time delay will occur in the estimation, and the accuracy of determining the current glucose value will also decrease.

In one embodiment of the invention, a lower bound of the variance is first determined for this purpose.
This corresponds to the minimum dispersion of the measurement system resulting from physical and technical considerations.
The initial value of the measurement variance is set based on the expected measurement variance.

これにより、測定値は、カルマン平滑器によってフィルタリングされる。
次いで、有意水準αでのコルモゴロー－スミルノー検定を使用して、サンプルが平均値

定分散σ_νが続く。 To determine the variance of the measurement noise values, the noise from a set of N measurements and associated filtered measurements is first calculated.

Thereby, the measured values are filtered by a Kalman smoother.
The Kolmogorow-Smirnow test at significance level α is then used to determine whether the sample has the mean value

followed by constant variance σ _ν .

となり、式中、ｄｆは、フィルタリングの自由度の数である。 The variance is now independent of the test result by the sample variance of the named samples,

where df is the number of degrees of freedom for filtering.

When the null hypothesis of the Kolmogorow-Smirnow test is rejected at the significance level α, the noise of the (next) N measurements is determined and the variance is replaced by the sample variance of the sample until the null hypothesis is no longer rejected. It will be done.

であり、

この目的のために、特に、有意水準αでの片側χ^２検定

が実行される（ｄｉｍ［ｚ］＝１およびα＝０，０５はｒ＝７，８８である）。 In another embodiment, outliers in the measurements are detected.
The so-called "squared normalized innovation value", or NIS value (consistency estimate) for short, can be used to detect outliers.
This tests "Normalized Innovation Squared" (NIS)

and

For this purpose, in particular, a one-sided ^χ2 test at significance level α

is executed (dim[z]=1 and α=0,05 with r=7,88).

Model errors, for example a strong increase in the case where elevated blood sugar also appears in the tissue signal, can also lead to exceeding the limit value, with the result that the measured value is incorrectly recognized as an outlier by the NIS test.
These values are then filtered through a Kalman filter, but this is not required.

Accordingly, in an embodiment of the invention, the measurement signal may also be checked such that a measurement value is identified as an outlier only if it falls below and/or exceeds a certain limit value.
Such outliers in the measurements are based on erroneous measurements caused, for example, by pressure fluctuations that become noticeable with large changes in temperature.

In one embodiment of the invention, several consecutive erroneous measurements may also be determined.

ｉ＝０；
ｃｏｎｔｉｎｕｅ；
ｅｌｓｅ
ｉ＋＋；
ｂｒｅａｋ； To identify these, the following procedure can be used.
if(i>1)
i=0;
break;
else if (measuredvalues(k)< lower_limit | measuredvalues(k)> upper_limit)

i=0;
continue;
else
i++;
break;

This prevents unnecessarily selecting measured values due to poor model description, or erroneously selecting measured values that do not satisfy the limit value condition but for which Kalman filter consistency is guaranteed.
This allows a restrictive selection of the respective limit values.
In other words, the number of measurements incorrectly classified as outliers due to model uncertainty can be reduced.

In one embodiment of the invention, the system malfunctions if the NIS outlier test is negative but at least two previous measurements are classified as measurement errors (i>1). A method is used in which it is assumed that the limit value is met and is considered exceeded only when the limit value is met.

In one embodiment of the invention, the current measurement value is filtered if the limit value condition is met, but at least two previous measurements are classified as measurement errors.
This allows relaxation processes to be taken into account.
However, the counter i according to the above method is set to 0 again.

In one embodiment of the invention, the number of arithmetic operations is reduced by approximating the BG-IG dynamics.
Since the covariance of the predictions is symmetric, it is sufficient to determine 6 out of 9 entries, which means that additional arithmetic operations can be saved.

In one embodiment of the invention, if the state transition matrix, the measurement matrix, and the covariance matrices of the process noise and measurement noise values are constant over a period of time, then in this case the covariance matrix of the prediction converges inversely. Therefore, calculation effort can be reduced.
If the change in the intermediate matrix elements is below a limit value, the algorithm is reduced to calculating state and measurement predictions and updated states.
In this way, only the state prediction step and the innovation step have to be performed step by step.

Here, an extended Kalman filter in the form of a so-called "Kalman fixed interval smoother" is used to filter the past measurements.

In addition to the possible benefits in calculating the rate of change ROC of blood glucose and improving the estimation results, filtering past values can also have the benefit of allowing the data to be used in a graphical display.
Depending on the measurement noise value of the Kalman filter, the blood glucose signal is relatively noisy, which is non-physiological (curve 101 in FIG. 2).
This can only be reduced by a more conservative setting of the parameters, ie a variance of the measurement noise values.
However, this introduces additional time delays.
However, by using a Kalman smoother, a smooth blood glucose signal (curve 102 in FIG. 2) can be provided without additional time delay.

記憶される。 The "Kalman fixed interval smoother" estimates the state at a fixed past interval of length T from the measurements of this interval.
They are based on the solution of the "Bayesian optimal smoothing" equation and consist of a two-pass filter, the forward pass corresponding to a Kalman filter.
Computation of the smoothed state at intervals of length T in the backward execution of the two-pass filter

be remembered.

In one embodiment of the invention, a Rauch-Tung-Striebel (RTS) smoothing filter may be used for this purpose.

に基づき、

は、予測の現在の推定値および共分散に対応する。 The RTS filter has the following filter formula calculated for all n-T<k<n, namely:

Based on

corresponds to the current estimate and covariance of the prediction.

ある。

be.

In further embodiments, an extended Rauch-Tung-Striebel (ERTS) smoother may be used for the nonlinear measurement model.

合、計算労力は、大幅に少なくなる。 In further embodiments, a Modified Bryson Frazier (MBF) smoother may be used.
It also uses the results of the Kalman filter in the forward run for smoothing in the backward run.

In this case, the computational effort is significantly less.

The calculation steps required for MBF filter are:

を使用して近似され、式中、

および

である。 In particular, MBF filters can be used with measurement models in the form of nonlinear functions.
The measurement model is a first-order Taylor polynomial

is approximated using, where,

and

It is.

Estimating blood sugar as accurately as possible involves calibration as well as evaluation of the data by a physician.
Optimization of blood glucose estimation by Kalman smoother reduces blood glucose overshoot and undershoot, which would otherwise significantly overestimate hyperglycemic and hypoglycemic states. Related to the analysis of blood sugar status (see Figure 3).

る。

らす。 When using self-monitoring of blood glucose data, i.e. blood glucose self-measurement SMBG to calibrate the whole system, accurate estimation of blood glucose values is also relevant, since estimation errors lead to poor identification of sensor parameters. do.
For example, for a one-point calibration using an estimate of the SMBG value and the corresponding blood glucose value.

Ru.

Ras.

One of the possible advantages of using the Kalman smoother online, i.e. using the Kalman smoother continuously compared to known post-processing of data, is computational efficiency.
The last T results of the Kalman filter are needed for the Kalman smoother.
In the case of post-processing using the Kalman smoother, these have to be re-determined or completely stored.
Other well-known methods such as normalized unfolding are significantly computationally expensive.

これは、非生理学的な値およびノイズの多い結果をもたらす可能性がある。
カルマン平滑器を使用した過去の値の改善された推定のおかげで、現在の血糖変化のより正確な推定を行うことができる。
血中グルコース値の変化ＲＯＣは、現在のＫＦ血中グルコース推定値とフィルタリングされたＫＳ血中グルコース値ｍサンプリングステップバックとの間の差から生じる。

For example, in the case of changes in blood sugar levels due to food intake or insulin, i.e. when the dynamic model can only roughly reproduce the time of change of state, larger estimation errors also result.
This data is used to estimate the average blood glucose change over the past mΔt=10 minutes.

This can lead to non-physiological values and noisy results.
Thanks to the improved estimation of past values using the Kalman smoother, a more accurate estimation of current blood glucose changes can be made.
The blood glucose value change ROC results from the difference between the current KF blood glucose estimate and the filtered KS blood glucose value m sampling step back.

Current blood sugar changes are visualized to the user by arrow symbols.
The division is done, for example, in groups of 5 or 7 (see table and FIG. 3).

Meanwhile, the recorded measurement values 200 are shown in FIG.
Comparison of blood glucose profile 201 using a Kalman filter and blood glucose profile 202 using a Kalman smoother allows for improved classification or improved trends by estimating the rate of change ROC using the filtered signal. (See reference numeral 203 in FIG. 3).
The upper part of the classification 203 is based on the blood sugar profile 201, the middle part of the classification 203 is based on the blood sugar profile 202, and the lower part is based on the measured value 201.
Measured value 201 represents the venous blood sugar level.
These are measured here using a YSI 2300 Stat Plus meter.
The circled arrow in stage 203 indicates a more accurate classification of trends using a comparison of Kalman smoothers and Kalman filters.

In summary, at least one embodiment of the invention provides at least one of the following advantages and/or features.
- modeling the diffusion process, using at least one Kalman filter in the case of a measurement model in the form of a linear function or using at least one extended Kalman filter in the case of a measurement model in the form of a non-linear function; Compensation for time delays by estimating current glucose values based on two provided state transition models and an approximation of the determined tissue glucose values.
- Robustness to outliers in the measured signal.
- Adaptation of slowly changing model parameters.
- Efficient implementation ensures high accuracy with low computational effort.
- Time and computationally efficient estimation of blood glucose.
- Adaptation of model parameters.
- Improving the robustness of estimation by introducing constraints.
- Flexibility regarding measurement models, eg being able to use both linear and non-linear sensor models and switching between them as required.
- Low computational effort to protect limited battery life.
- Improved system calibration based on SMGB measurements.

Although the present invention has been described above based on the embodiments, the present invention is not limited to these and can be modified in various ways.

101... Blood glucose Kalman filter 102... Blood sugar Kalman smoother 200... Read measured value 201... Progress of blood glucose value using Kalman filter 202... Blood glucose level using Kalman smoother Glucose value progress 203...Classification/trend S1-S4...Method steps

Claims (18)

A method for continuously estimating current glucose values in a transport fluid of a living organism, in particular blood, comprising:
a) measuring a series of measurements comprising at least two measurements using a sensor for temporally spaced tissue glucose values in tissue surrounding the transport fluid;
b) determining the tissue glucose value from a series of measurements of the sensor, taking into account measurement noise values, based on a measurement model in the form of a linear or non-linear function;
c) providing (S3) a state transition model used to assign the current glucose value in the transport fluid to the determined tissue glucose value taking into account process noise values;
d) said provided state transition model and said determined tissue using a Kalman filter in the case of a measurement model in the form of a linear function or using an extended Kalman filter in the case of a measurement model in the form of a non-linear function. estimating the current glucose value in the transport fluid based on an approximation of the glucose value;
method including. Method according to claim 1, characterized in that several state transition models are provided that vary depending on the course of the estimated current glucose value over time, in particular its rate of change over time. . At least two state transition models are provided, one based on a constant glucose concentration, one based on a constant change in glucose concentration, and/or one based on a weighted sum of previous glucose concentrations. 3. A method according to claim 1 or claim 2, characterized in that: the measured value is filtered by at least one filter function;
4. Method according to any one of claims 1 to 3, characterized in that errors of the sensor, in particular measurement errors, are suppressed by the filter function. 5. Method according to any one of claims 1 to 4, characterized in that the measurement noise value is adjusted periodically. 6. Method according to claim 5, characterized in that the variance of the measurement noise value is determined, in particular estimated, using a random sample of the measurement values for adapting the measurement noise value. A statistical test, in particular a Kolmogorov-Smirnov test, is used to check whether the null hypothesis (that the sample follows a mean free Gaussian distribution with the determined variance of the measured noise values) is not rejected. 7. Method according to claim 6, characterized in that. 8. Method according to claim 7, characterized in that the variance of the measurement noise values is determined for further samples of measurement values as long as the null hypothesis is rejected. 5. The method according to claim 4, wherein the filter function is used to check the measurements for outliers using a NIS test, and measurements determined as outliers are discarded. Method. 4. The filter function is used to check the measured values to determine whether they exceed or fall below a specified limit value, after which measured values above or below the limit value are discarded. 9. If a predetermined number, in particular two temporally consecutive previous measurements, were previously rejected as measurement errors, then the current measurement value, which was not determined as an outlier, is nevertheless rejected as a measurement error. 11. The method according to claim 10, characterized in that: 12. Any one of claims 1 to 11, characterized in that the state transition model comprises a diffusion model for time-dependent modeling of the diffusion process of glucose from the transport fluid into the surrounding tissue. The method described in. 13. Method according to any one of claims 1 to 12, characterized in that several previous measurements are filtered by a Kalman fixed interval smoother. When the Kalman fixed interval smoother is used, it is performed in forward and reverse directions, characterized in that in the forward pass Kalman filtering is used and in the reverse pass an RTS filter and/or an MBF filter is used. 14. The method according to claim 13. 15. A method according to any one of claims 1 to 14, characterized in that the trends in blood glucose concentration are classified using categories, using at least seven categories. Apparatus, in particular for carrying out the method according to any one of claims 1 to 15, for continuously estimating the current glucose value in a transport fluid of a living organism, in particular blood, comprising:
A sensor for measuring fluorescence in said transport fluid surrounding tissue by means of a probe, in particular a polymeric fiber optic probe, designed to determine a series of measurements, said sensor for measuring fluorescence in said transport fluid surrounding said transport fluid; a sensor comprising at least two temporally spaced measurements for a tissue glucose value;
A providing apparatus designed to provide a state transition model for assigning a glucose value in the transport fluid to the determined tissue glucose value taking into account a process noise value. , also providing a measurement model in the form of a linear or non-linear function, whereby the measurement value of the sensor is used by the measurement model to be assigned to the tissue glucose value, taking into account measurement noise values. a providing device;
Based on the measurement model, the series of measurements is used to determine the tissue glucose value, and using a Kalman filter in the case of a measurement model in the form of a linear function, or using a Kalman filter in the case of a measurement model in the form of a linear function; designed to estimate the current glucose value in the transport fluid based on a provided state transition model and an approximation of the tissue glucose value using an extended Kalman filter in the case of a morphology measurement model; An evaluation device; An evaluation device for continuously estimating the current glucose value in a transport fluid of a living organism, in particular blood, comprising:
an interface for connecting a sensor for providing a series of measurements, the interface comprising at least two measurements at different times of tissue glucose values in tissue surrounding the transport fluid;
a memory for storing a state transition model, the state transition model converting the tissue glucose value determined by the state transition model into a glucose value in the transport fluid, taking into account a process noise value; assignment, is used to store a measurement model in the form of a linear or non-linear function, said measurement model being used to assign said sensor measurements to said tissue glucose values, taking into account measurement noise values; memory and
determining the tissue glucose value using the series of measurements based on the stored measurement model and, in the case of a measurement model in the form of a linear function, using a Kalman filter; or designed to estimate the current glucose value in the transport fluid based on the state transition model and an approximation of the tissue glucose value using an extended Kalman filter in the case of a measurement model in the form of a non-linear function; an evaluation device including: 16. Any one of claims 1 to 15 for storing instructions which, when executed on a computer, cause the method to perform a method for continuously estimating the current glucose value in a transport fluid of a living organism, in particular blood. a non-volatile computer-readable medium suitable for carrying out the method described in
a) measuring with a sensor a series of measurements comprising at least two temporally spaced measurements of tissue glucose values in tissue surrounding the transport fluid;
b) determining the tissue glucose value using the series of measurements based on a measurement model in the form of a linear or non-linear function, the measurement model taking into account measurement noise values; used to assign the sensor reading to a tissue glucose value;
c) providing a state transition model, wherein the state transition model uses the determined tissue glucose value, taking into account process noise values, and wherein the glucose value in the transport fluid is a tissue glucose value. a step that is assigned to
d) based on the approximation of said state transition model and said tissue glucose value using a Kalman filter in the case of a measurement model in the form of a linear function or using an extended Kalman filter in the case of a measurement model in the form of a non-linear function; estimating the current glucose value in the transport fluid.

Images